What is the probability that a sequence that folds into a high scoring shape could also fold into a low scoring shape? And vice versa? Is there a statistical curve or something similar available that suggests how repeatable our results are expected to be?
Good question. For some of the designs we have carried out replicate experiments and seen that the ‘eterna score’ is reproduced with a mean unsigned error +/-4% – that is 3 or 4 nts in a 100 nucleotide design can change a bit.
However, another idea that might more directly answer your question would be to re-model each RNA, guided by the lab chemical mapping data, with ‘bootstrapping’ to estimate how confidently each stem is recovered in the structure, compared to all other structure. This procedure is described here.
We haven’t implemented it in the game (its a bit computationally intensive), but perhaps we could give it a shot in the future.